An overview of free software tools for general

data mining
A. Jovi*, K. Brki* and N. Bogunovi*
*

Faculty of Electrical Engineering and Computing, University of Zagreb / Department of Electronics, Microelectronics,
Computer and Intelligent Systems, Unska 3, 10 000 Zagreb, Croatia
{alan.jovic, karla.brkic, nikola.bogunovic}@fer.hr

Abstract - This expert paper describes the characteristics of

six most used free software tools for general data mining
that are available today: RapidMiner, R, Weka, KNIME,
Orange, and scikit-learn. The goal is to provide the
interested researcher with all the important pros and cons
regarding the use of a particular tool. A comparison of the
implemented algorithms covering all areas of data mining
(classification, regression, clustering, associative rules,
feature selection, evaluation criteria, visualization, etc.) is
provided. In addition, the tools' support for the more
advanced and specialized research topics (big data, data
streams, text mining, etc.) is outlined, where applicable. The
tools are also compared with respect to the community
support, based on the available sources. This
multidimensional overview in the form of expert paper on
data mining tools emphasizes the quality of RapidMiner, R,
Weka, and KNIME platforms, but also acknowledges the
significant advancements made in the other tools.

I.

INTRODUCTION

Data mining (DM) is the core step in knowledge

discovery in datasets. It integrates all the analysis
procedures that are required in order to reveal new and
relevant information to an interested user. DM includes
data preparation and data modeling. Datasets may be
obtained from a variety of sources, including: traditional
relational databases, data warehouses, web documents, or
simple local textual files. It is important to prepare data in
the most efficient way in order to extract as much
information as possible [1]. After preparation, various
models can be constructed, depending on the research
goal. In order to properly interpret the models, standard
evaluation and statistical procedures are pursued. The
visualization of the results is also encouraged.
Free and publicly available software tools for DM
have been in development for the past 20 years. The goal
of these tools is to facilitate the rather complicated data
analysis process and to offer all interested researchers a
free alternative to commercial data analysis platforms.
They do so mainly by proposing integrated environments
or specialized packages on top of standard programming
languages, which are often open source.
This paper focuses on the review of several such tools
that have grown more efficient and useful over the years,
some even comparable or better in certain aspects than
their commercial counterparts. In particular, the main
characteristics of RapidMiner [2], R [3], Weka [4],

Orange [5], KNIME [6], and scikit-learn [7] will be

outlined and compared. All of these free tools have
implementations of general DM tasks. Other, more
specialized tools such as Elki (clustering) or Anatella (big
data), and tools with small DM community support (e.g.
F#, GNU Octave) are not considered here.
The tools are compared in terms of general properties
(language, license, etc.), core DM tasks (supported input,
data preparation, data analysis, output), and support for
some recent more advanced DM topics (big data, data
streams, text mining, deep learning).
II.

FREE DATA MINING TOOLS OVERVIEW

In a recent, 2013, poll published on the influential

KDnuggets portal [8], regarding the use of DM tools in a
real project, it is interesting to observe that in the top 5
tools there is only one commercial tool: Excel. The
domination of free tools, mainly RapidMiner and R,
probably stems from the maturity and availability of a
large number of machine learning algorithm
implementations. In Fig. 1, an adapted version of the poll
is shown, with only free tools listed, for comparison.
Most of the modern DM tools are effectively softwarebased dataflow architectures. Some of the tools (e.g.
RapidMiner, KNIME) are graphical integrated
environments that enable visual component placement,
connection, and dragging. The most common paradigm
for such components is: pull the data in, transform, and
push it further down the pipeline. The component only
pulls the data in once all of its prerequisites are met. The
under-the-hood implementation of such components is
irrelevant for the average user, but more advanced users
usually have the possibility of tackling with the code in
order to improve it. Other tools (e.g. R) are just plain
extensions of the underlying language in the form of
specialized packages and/or GUI add-ons.
General characteristics of the six DM tools are listed in
Table I. All of the tools have implementations for
Windows, Linux, and Mac OS X operating systems. In the
large Table II, the supported machine learning algorithms
for each DM tool are summarized. The tools either
implement an algorithm (+), use an external add-on (A) to
support it, show some degree of support for the procedure
(S), or do not implement it (-) at all. It must be noted here
that the data in Table II should be considered temporary,
because most of the tools are in constant state of upgrades.
Nevertheless, it is important and useful to summarize their

TABLE I.
Characteristic
Developer:
Programming
language:
License:
Current
version:
GUI /
command line:

GENERAL CHARACTERISTICS OF THE FREE DATA MINING TOOLS

RapidMiner
RapidMiner,
Germany

R
worldwide
development

Weka
Univ. of Waikato,
New Zealand

Java

C, Fortran, R

Java

open s. (v.5 or
lower); closed s.,
free Starter ed. (v.6)

free software,
GNU GPL 2+

6
GUI

Main purpose:

general data mining

Community
support (est.):

large
(~200 000 users)

Orange
Univ. of Ljubljana,
Slovenia
C++, Python,
Qt framew.

KNIME
KNIME.com
AG,Switzerland

open source,
GNU GPL 3

open source,
GNU GPL 3

open source,
GNU GPL 3

FreeBSD

3.02

3.6.10

2.7

2.9.1

0.14.1

both; (GUI for

DM = Rattle)
sci. computation
and statistics
very large
(~ 2 M users)

both

both

GUI

command line

general data
mining

general data
mining

machine learning
package add-on

large

moderate

general data
mining
moderate
(~ 15 000 users)

Java

scikit-learn
multiple; support:
INRIA, Google
Python+NumPy+
SciPy+matplotlib

moderate

the company RapidMiner, Germany. Previous versions (v.

5 or lower) were open source. The latest one (v. 6) is
proprietary for now, with several license options (Starter,
Personal, Professional, Enterprise). The Starter version is
free with limitations only in respect to maximum allocated
memory size (1 GB) and input files (.csv, Excel). The tool
has become very popular in several recent years and has a
large community support.
RapidMiner offers an integrating environment with
visually appealing and user-friendly GUI. Everything in
RapidMiner is focused on processes that may contain subprocesses. Processes contain operators in the form of
visual components. Operators are implementations of DM
algorithms, data sources, and data sinks. The dataflow is
constructed by drag-and-drop of operators and by
connecting the inputs and outputs of corresponding
operators. RapidMiner also offers the option of
application wizards that construct the process
automatically based on the required project goals (e.g.
direct marketing, churn analysis, sentiment analysis).
There are tutorials available for many specific tasks so the
tool has a stable learning curve.

Figure 1. The community use of free DM tools, adapted from [8]

capabilities so that interested users can choose the

appropriate environment for handling their problem. The
algorithms shown in Table II were selected based on their
significance and presence in most of the tools. Although
the tools may contain some additional algorithms, not all
of them could have been listed. Some of these were put in
the appropriate rows with label others. References to the
work that describe the algorithms are not provided.
Table III lists the support for some of the more
specialized topics in DM such as big data, text mining,
etc.
III.

TOOLS DESCRIPTION

A. RapidMiner
RapidMiner (previously: Rapid-I, YALE) is a mature,
Java-based, general DM tool currently in development by

Although RapidMiner is quite powerful with its basic

set of operators, it is the extensions that make it even more
useful. Popular extensions include sets of operators for
text mining, web mining, time series analysis, etc. Most of
the operators from Weka are also available through
extension, which increases the number of implemented
DM methods (Table II). The tool has very few
shortcomings. The most important one is the transition to
a novel model of business. It remains to be seen whether
the transition to proprietary license will limit the number
of its users, but it may not be helpful. The support for deep
learning methods and some of the more advanced specific
machine learning algorithms (e.g. extremely randomized
trees, various inductive logic programming algorithms) is
currently limited. However, big data analysis via Hadoop
cluster (Radoop) is supported.
B. Weka
Weka is a Java-based, open-source DM platform
developed at the University of Waikato, New Zealand.
The software is free under GNU GPL 3 for noncommercial purposes. Weka has had mostly stable

TABLE II.
Category

Data import

Name
textual files (.txt,
.csv)
specific input format
files
Excel/
spreadsheet

DATA MINING ALGORITHMS AND PROCEDURES SUPPORTED BY THE TOOLS

RapidMiner

Weka

Orange

KNIME

+ (e.g..arff, .xrff)

A (foreign)

+ (.arff, .libsvm)

A (xlsx)

database table

A (RODBC)

+ (prototype)

data from an URL

filters

A (FSelector)

wrappers

A (FSelector)

discretization

A (RWeka)

normalization

A (RWeka)

PCA

ICA

+ (fastICA)

MDS

SVD

random projections

ID3

A (Weka)

A (Weka)

C4.5

A (Weka)

A (RWeka)

CART

A (Weka)

A (RWeka)

A (Weka)

others

+, A (own*, dec.
stump)

+, A (own*,
RWeka)

+ (dec. stump)

+ (own*)

+ (own*)

1Rule

A (RWeka)

A (Weka)

PART

A (Weka)

A (RWeka)

A (Weka)

A (RWeka)

A (Weka)
A (Weka)

Feature selection

Feature
transformation

Decision tree
learner

Classification
rules

RIPPER

+ (subgroup
discovery)

A (RWeka)

+ (Ridor)

+ (CN2, subgroup
discovery)

full bayesian network

A (Weka)

A (Weka)

AODE

A (Weka)

A (AnDE)

A (Weka)

A (Weka)

others
Naive Bayes
Bayesian
networks

Instance based
learning

others

A (Weka)

+ (DMNB,
WAODE)

kNN

A (class,
RWeka)

LWL

A (Weka)

A (stats)

A (Weka)

A (Weka)
e.g. LBR
+ (log., lin.,
polynomial)

A (RWeka)
e.g. LBR

+ (LBR)

+ (ML-kNN)

A (Weka)

+ (log, lin.)

+ (log., lin., Lasso,

PLS, trees, mean)

+(log., lin.,
poly, trees)

others
regression analysis
Function based
learning

ANN
SVM
others

Hybrid learning
methods

Ensemble

+ (MLP)
+ (linear, evolut.,
PSO)
+ (Gaussian proc.,
RVM)

+ (lin., nonlin.)
A (nnet,
RSNNS)
A (e1071,
RWeka)

+ (MLP)

MLP

MLP, PNN

+ (SMO), A
(libsvm)

+ (libsvm,
liblinear)

+ (integr.
libSVM)

logistic model trees

A (Weka)

A (RWeka)

A (Weka)

function trees

A (Weka)

A (Weka)

Bayesian trees

A (Weka)

A (Weka)

A (Weka)

A (Weka)

others

+ (DTNB), A
(CBA)

bagging

A (RWeka)

Category
learning

Name

RapidMiner

Weka

Orange

KNIME

A (RWeka)

A (Weka)

random forest

A (random
Forest)

extremely
randomized trees

+ (extra Trees)

A (Weka)

A (Weka)

LogitBoost

A (RWeka)

A (Weka)

option tree

A (Weka)

A (Weka)

AdaBoost

rotation forest

stacking

A (RWeka)

A (Weka)

+ (Bayesian
boosting)

+ (boosted
regression)

+ (Multi
Boost)

A (Weka)

linkage-based

AGNES

A (cluster)

BIRCH

A (birch)

k-means

X-means

A (RWeka)

A (Weka)

A (cluster)

A (e1071)

EM clustering

A (mclust)

A (Weka)

DBSCAN

A (fpc)

A (Weka)

A (Weka)

A (Weka)

Self-organising map

A(RSNNS)

A (Weka)

GSP

A (Weka)

A (Weka)

A (arules,
RWeka)

+ (sparse, attr.value)

A (Weka)

FP-growth

A (Weka)

Eclat

A (arules)

A (Weka)

A (RWeka)

A (Weka)

others

A (arules
Sequences)

+ (predictive
Apriori)

+ (Apriori-SD)

A (Weka)

holdout

cross-validation

A (cvTools)

A (ROCR)

A (ROCR)

A (pROC)

histograms

scatterplots

A (lattice)

S
(scatterplot3D)

other

Hierarchical
clustering

Centroid
(partition) based
clustering

Distribution based
clustering
Density based
clustering
ANN based
clustering

Partition Around
Medoids
fuzzy c-means
clustering

OPTICS

Apriori
Association rules
(unsupervised)

Tertius

Evaluation
methods and
metrics

Data visualization

classif. accur., mean

abs.err., MSE,
TP, FP, FN, TN conf.
matrix, precis., recall
ROC, Lift chart,
Cost-benefit

other plots (e.g. boxwhisker, mean/error)

3D graphs (e.g. bivar.
hist., surface plots)

* Own implementation of decision tree, mostly similar to C4.5 and CART

popularity over the years, which is mainly due to its user

friendliness and the availability of a large number of
implemented DM algorithms. It is still not as popular as
RapidMiner or R, both in business and academic circles,
mostly because of some slow and more resource
demanding implementations of DM algorithms. Although

it is not a single tool of choice in DM, Weka is still quite

powerful and versatile, and has a large community
support.
Weka offers four options for DM: command-line
interface (CLI), Explorer, Experimenter, and Knowledge
flow. The preferred option is the Explorer which allows

TABLE III.
Name
Big data
Link, graph
mining
Spatial data
analysis
Time-series
analysis
Semi-super-vised
learning

RapidMiner
S (not free:
Radoop)

SUPPORT FOR SPECIALIZED AND ADVANCED DATA MINING TASKS

R
A (ff, ffbase)

Weka
S (CLI, knowl. flow,
distributedWekaHadoop)

Orange

KNIME

scikit-learn

A (igraph, sna)

A (ggmap)

+, A(forecast)

S (several time series filters)

A (upclass)

A (stream)

A (massiveOnlineAnalysis)

Text mining

A (tm, RTextTools,
qdap)

Paralelization

S (enterprise
ed.)

A (snow, multicore)

A (joblib)

S (darch:
incomplete)

S (Restricted
Boltzmann Mach.)

Data streams

Deep learning

the definition of data source, data preparation, machine

learning algorithms, and visualization. The Experimenter
is used mainly for comparison of the performance of
different algorithms on the same dataset. Knowledge flow
is akin to RapidMiner's operator paradigm in a sense that
it allows one to specify the dataflow using appropriately
connected visual components. Although not as visually
appealing and extendable as RapidMiner, Weka's
Knowledge flow still does the job.
Weka supports many model evaluation procedures and
metrics, but lacks many data survey and visualization
methods. It is also more oriented towards classification
and regression problems and less towards descriptive
statistics and clustering methods, although some
improvements were made recently with respect to
clustering. The support for big data, text mining, and
semi-supervised learning is currently limited, while deep
learning methods are still not considered.
C. R
The open-source tool and programming language of
choice for statisticians, R, is also a strong option for DM
tasks. R has been in development for the last 15 years and
is the successor of S, a statistical language originally
developed by Bell Labs in 1970s. The source code of R is
written in C++, Fortran, and in R itself. It is an interpreted
language and is mostly optimized for matrix based
calculations, comparable in performance to commercially
available Matlab and freely available GNU Octave.
The main language is extended by a myriad collection
of packages for all sorts of computational tasks. Some of
the packages are listed in Table II in parentheses. The tool
offers only a simple GUI with command-line shell for
input. It is certainly not a user-friendly environment
because all commands need to be entered in the R
language. The learning curve is steep, and although simple
tasks such as drawing graphs and descriptive statistics can
be learned rather easy, the language's full potential is
difficult to master.
Some advanced users sometimes offer helpful
reference cards with the list of the most significant

S (timeseries
module has bugs)
+ (label
propagation)

functions [9]. From DM user's perspective, R offers very

fast implementations of many machine learning
algorithms, comparable in number to RapidMiner and
Weka (from which a large number of algorithms is
borrowed), and also the full prospect of statistical data
visualizations methods. It has specific data types for
handling big data, supports parallelization, data streams,
web mining, graph mining, spatial mining, and many other
advanced tasks, including a limited support for deep
learning methods.
R's main problem is its language, which, although
highly extendable, is also a difficult one to learn
thoroughly enough to become productive in DM.
Advancement for DM tasks in that direction is the Rattle
package (author Dr. Graham Williams and other
contributors) that offers a decent GUI for R. Rattle, which
is in development from 2006, is similar to Weka's
Explorer in the sense of user-friendliness. It loads separate
packages from R upon request for a specific analysis.
Rattle uses some of the R's standard implementations of
DM methods and also additional packages. The only
problem with Rattle is that it cannot use all of the R's
algorithms or Weka's DM implementations. Nevertheless,
Rattle is user-friendly and quite popular in DM
community.
D. Orange
Orange is a Python-based tool for DM being
developed at the Bioinformatics Laboratory of the Faculty
of Computer and Information Science at the University of
Ljubljana. It can be used either through Python scripting
as a Python plug-in, or through visual programming. Its
visual programming interface, Orange Canvas, offers a
structured view of supported functionalities grouped into
nine
categories: data operations,
visualization,
classification, regression, evaluation, unsupervised
learning, association, visualization using Qt, and prototype
implementations.
Functionalities are visually represented by different
widgets (e.g. read file, discretize, train SVM classifier
etc). A short description of each widget is available within
the interface. Programming is performed by placing

widgets on the canvas and connecting their inputs and

outputs. The interface is very polished and visually
appealing, offering a pleasant user experience.
One apparent downside of Orange is that the number
of available widgets seems limited when compared to
other tools such as RapidMiner or KNIME, especially
because of the lack of integration with Weka. Still, the
coverage of standard data mining techniques is quite good,
as can be seen from Table II. Furthermore, there are a
number of interesting widgets currently in development
that can be found in the "Prototype" category, so it is
reasonable to expect that the feature set will be expanded
in the future.
E. scikit-learn
scikit-learn is a free package in Python that extends the
functionality of NumPy and SciPy packages with
numerous DM algorithms. It also uses the matplotlib
package for plotting charts. The package keeps improving
by accepting valuable contributions from many
contributors and is supported by both INRIA and Google
Summer of Code. One of its main strong points is a wellwritten online documentation for all of the implemented
algorithms. Well-written documentation is a requirement
for any contributor and is valued more than a lot of poorly
documented algorithm implementations.
The package supports most of the core DM
algorithms. However, several significant DM algorithm
groups have been omitted currently, including
classification rules and association rules. On the other
hand, the package is strong in function-based methods
including many general linear models and various types of
SVM implementations. It is also quite fast despite being
written in an interpreted language. This is mainly because
the contributors are asked to optimize the code in various
aspects, such as calling array based NumPy number
crunching algorithm or writing wrappers for existing
C/C++ implementations in Cython.
Despite its advantages, the use of scikit-learn still
requires that one is a skilled programmer in Python
because of its command-line interface. This will detract
almost anyone not versed in this language because there
are other tools that do not have this assumption.

1000 nodes are available through the core installation and

various extensions. Nodes are organized in a hierarchy
and can be searched by name within an intuitive interface.
Each node is documented in detail, and the documentation
is automatically shown within the interface once the node
is selected.
A large repository of example workflows is available
to facilitate quicker learning of the tool. One of the
greatest strengths of KNIME is the integration with Weka
and R. Although extensions have to be installed to enable
the integration, the installation itself is trivial. Weka
integration enables using almost all the functionality
available in Weka as KNIME nodes, while R integration
enables running R code as a step in the workflow, opening
R views and learning models within R. Several other
interesting free extensions are also available, e.g.
JFreeChart extension that enables advanced charting,
OpenStreetMap extension that enables working with
geographical data, etc. There are also commercial
extensions for more specific functionalities. Overall,
KNIME seems to be one of the best choices for a user
interested in a purely visual programming paradigm with a
need for a large variety of nodes.
IV.

Several DM tools were presented in this work. Overall

conclusion is that there is no single best tool. Each tool
has its strong points and weaknesses. Nevertheless,
RapidMiner, R, Weka, and KNIME have most of the
desired characteristics for a fully-functional DM platform
and therefore their use can be recommended for most of
the DM tasks.
REFERENCES
[1]
[2]

[3]
[4]

[5]

F. KNIME
KNIME (Konstanz Information Miner) is a generalpurpose DM tool based on the Eclipse platform,
developed and maintained by the Swiss company
KNIME.com AG. Its development started in 2004 at the
University of Konstanz, Germany, and the initial version
was released in 2006. KNIME is open-source, though
commercial licenses exist for companies requiring
professional technical support. According to the official
website, KNIME is used by over 3000 organizations in
more than 60 countries, and there seems to be a
considerable community support.
The tool adheres to the visual programming paradigm
present in most DM tools, where building blocks are
placed on a canvas and connected to obtain a visual
program. In KNIME, these building blocks are called
nodes, and according to the official website, more than

CONCLUSION

[6]

[7]

[8]

[9]

D. Pyle, Data Preparation for Data Mining, San Diego: Academic

Press, 1999.
M. Hofmann and R. Klinkenberg, RapidMiner: Data Mining Use
Cases and Business Analytics Applications, Boca Raton: CRC
Press, 2013.
Y. Zhao, R and Data Mining: Examples and Case Studies, San
Diego: Academic Press, 2012.
M. Hall, E. Frank, G. Holmes, B. Pfahringer, P. Reutemann, and
I. H. Witten, The WEKA data mining software: an update,
SIGKDD Explorations, vol. 11, no. 1, pp. 1018, 2009.
J. Demar, T. Curk, and A. Erjavec, Orange: Data Mining
Toolbox in Python, Journal of Machine Learning Research, vol.
14, pp. 23492353, 2013.
M. R. Berthold, N. Cebron, F. Dill, T. R. Gabriel, T. Ktter, T.
Meinl, et al., KNIME: The Konstanz Information Miner, in Data
Analysis, Machine Learning and Applications (Studies in
Classification, Data Analysis, and Knowledge Organization),
Springer Berlin Heidelberg, pp. 319326, 2008.
F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion,
O. Grisel, et al., Scikit-learn: Machine Learning in Python,
Journal of Machine Learning Research, vol. 12, pp. 28252830,
2011.
G. Piatetsky, KDnuggets Annual Software Poll: RapidMiner and
R vie for first place, 2013, Available at [last accessed 2014-0222]:
http://www.kdnuggets.com/2013/06/kdnuggets-annualsoftware-poll-rapidminer-r-vie-for-first-place.html
Y. Zhao, R Reference Card for Data Mining, Available at: [last
accessed
2014-02-23]
http://www.rdatamining.com/docs/Rrefcard-data-mining.pdf